Summarizing User Opinions: A Method for Labeled-data Scarce Product Domains

Product reviews contain valuable information that can influence the online purchases. Extracting relevant opinions regarding the product by merely reading all the reviews is a herculean task. An automatic method for mining and summarizing opinions in these reviews is necessary for this purpose. Existing methods for opinion summarization requires pre-labeled data from the target domain or other sophisticated lexical resources. We solve the problems of existing methods by using cross-domain sentiment classification coupled with distributional similarity of opinion words to classify and summarize product reviews. Experimental analysis shows that using cross-domain sentiment classification for opinion summarization gives encouraging results.